1_3_f2py.ipynb

{
 "metadata": {
  "name": "",
  "signature": "sha256:e64a7a2f12e3018d89773a032583cef8ae5adcff62101d721e4238e97d8f8b95"
 },
 "nbformat": 3,
 "nbformat_minor": 0,
 "worksheets": [
  {
   "cells": [
    {
     "cell_type": "markdown",
     "metadata": {},
     "source": [
      "#Glue compiled libraries to Python#\n",
      "\n",
      "How to speed up loop ?"
     ]
    },
    {
     "cell_type": "markdown",
     "metadata": {},
     "source": [
      "##Ctypes##\n",
      "Ctypes, available from Python's standart library, provides C compatible data types, and allows calling functions or shared libraries. \n",
      "It can be used to wrap these libraries in pure Python... but requires (some) additionnal work\n",
      "\n",
      "Numpy's array have a ctypes attribute which contains the address of the underlying buffer. \n",
      "It is the user's responsiblity to ensure data are contiguous and properly alligned !"
     ]
    },
    {
     "cell_type": "code",
     "collapsed": false,
     "input": [
      "import numpy\n",
      "a = numpy.arange(10)\n",
      "print(a)\n",
      "print(a.strides)\n",
      "print(\"Address of the buffer: %s\" % a.ctypes.data)\n",
      "print(a.flags)\n",
      "b=a[::2]\n",
      "print(b)\n",
      "print(b.strides)\n",
      "print(\"Address of the buffer: %s\" % b.ctypes.data)\n",
      "print(b.flags)\n"
     ],
     "language": "python",
     "metadata": {},
     "outputs": [
      {
       "output_type": "stream",
       "stream": "stdout",
       "text": [
        "[0 1 2 3 4 5 6 7 8 9]\n",
        "(8,)\n",
        "Address of the buffer: 40848944\n",
        "  C_CONTIGUOUS : True\n",
        "  F_CONTIGUOUS : True\n",
        "  OWNDATA : True\n",
        "  WRITEABLE : True\n",
        "  ALIGNED : True\n",
        "  UPDATEIFCOPY : False\n",
        "[0 2 4 6 8]\n",
        "(16,)\n",
        "Address of the buffer: 40848944\n",
        "  C_CONTIGUOUS : False\n",
        "  F_CONTIGUOUS : False\n",
        "  OWNDATA : False\n",
        "  WRITEABLE : True\n",
        "  ALIGNED : True\n",
        "  UPDATEIFCOPY : False\n"
       ]
      }
     ],
     "prompt_number": 1
    },
    {
     "cell_type": "markdown",
     "metadata": {},
     "source": [
      " ##F2Py#\n",
      "Fortran to Python interface generator, provided by NumPy, which allows:\n",
      "\n",
      "* Calling Fortran 77/90/95, Fortran 90/95 module, and C functions from Python\n",
      "* Accessing Fortran 77 COMMON blocks and Fortran 90/95 module data from Python\n",
      "* Call-back Python functions from Fortran or C\n",
      "* Automatically handling the difference in the data storage order of multi-dimensional Fortran and Numerical Python (i.e. C) arrays.\n",
      "\n",
      "Of course F2Py needs a ForTran Compiler to compile Fortran code which can be an issue on some platforms."
     ]
    },
    {
     "cell_type": "code",
     "collapsed": false,
     "input": [
      "src = \"\"\"\n",
      "C FILE: FIB3.F\n",
      "      SUBROUTINE FIB(A,N)\n",
      "C\n",
      "C     CALCULATE FIRST N FIBONACCI NUMBERS\n",
      "C\n",
      "      INTEGER N\n",
      "      INTEGER A(N)\n",
      "Cf2py intent(in) n\n",
      "Cf2py intent(out) a\n",
      "Cf2py depend(n) a\n",
      "      DO I=1,N\n",
      "         IF (I.EQ.1) THEN\n",
      "            A(I) = 0\n",
      "         ELSEIF (I.EQ.2) THEN\n",
      "            A(I) = 1\n",
      "         ELSE \n",
      "            A(I) = A(I-1) + A(I-2)\n",
      "         ENDIF\n",
      "      ENDDO\n",
      "      END\n",
      "C END FILE FIB3.F\"\"\""
     ],
     "language": "python",
     "metadata": {},
     "outputs": [],
     "prompt_number": 2
    },
    {
     "cell_type": "markdown",
     "metadata": {},
     "source": [
      "Note:\n",
      "Fortran has no \"functions\" only \"subroutines\" (or procedure), so all output needs to be given by calling module.\n",
      "\n",
      "The 3 comment lines staring with Cf2py declare which variable are input, and which are output  "
     ]
    },
    {
     "cell_type": "code",
     "collapsed": false,
     "input": [
      "from numpy import f2py"
     ],
     "language": "python",
     "metadata": {},
     "outputs": [],
     "prompt_number": 3
    },
    {
     "cell_type": "code",
     "collapsed": false,
     "input": [
      "f2py.compile(src, \"fibo\")"
     ],
     "language": "python",
     "metadata": {},
     "outputs": [
      {
       "metadata": {},
       "output_type": "pyout",
       "prompt_number": 4,
       "text": [
        "0"
       ]
      }
     ],
     "prompt_number": 4
    },
    {
     "cell_type": "code",
     "collapsed": false,
     "input": [
      "import fibo\n",
      "#reload(fibo)"
     ],
     "language": "python",
     "metadata": {},
     "outputs": [],
     "prompt_number": 5
    },
    {
     "cell_type": "code",
     "collapsed": false,
     "input": [
      "fibo.fib(19)"
     ],
     "language": "python",
     "metadata": {},
     "outputs": [
      {
       "metadata": {},
       "output_type": "pyout",
       "prompt_number": 6,
       "text": [
        "array([   0,    1,    1,    2,    3,    5,    8,   13,   21,   34,   55,\n",
        "         89,  144,  233,  377,  610,  987, 1597, 2584], dtype=int32)"
       ]
      }
     ],
     "prompt_number": 6
    },
    {
     "cell_type": "markdown",
     "metadata": {},
     "source": [
      "#Other binary modules interfaces#\n",
      "##Weave##\n",
      "\n",
      "Weave is part of SciPy, it is a runtime compiler of C/C++ code to make loop go fast but it is deprecated now and remains only for compatibility reasons.\n",
      "While working effectively under UNIX, weave is had to set-up on windows computer (due to the absence of any kind of compiler by default).\n",
      "\n",
      "##Cython##\n",
      "\n",
      "As Python is written in C, any line of Python can be translated to its equivalent C part, using metaprogramming.\n",
      "The Pyrex project aimed at infering types to make the generated C-code easier to optimize for the compiler.\n",
      "Cython is the continuation of the Pyrex project with support for NumPy nd-arrays.\n",
      "\n",
      "Cython is the weave killer: many project replaced their hand-written binding or C code by Cython code.\n",
      "Used by LXML and most of the scikits\n",
      "\n",
      "##Boost-Python##\n",
      "C++ binding for python from the famous C++ boost library. Very large but also very efficient.\n",
      "It is used by many projects: PyOpenCL, PyCuda.\n",
      "\n",
      "##SWIG##\n",
      "General purpose binding for any kind of interpreted programming language\n",
      "\n",
      "##SIP##\n",
      "The Python-C++ binding from PyQt.\n",
      "\n",
      "##Shibokken##\n",
      "The Python-C++ binding from PySide.\n",
      "\n",
      "##CFFI##\n",
      "The Python-C interface invented by PyPy\n"
     ]
    },
    {
     "cell_type": "markdown",
     "metadata": {},
     "source": [
      "\n",
      "Well number crunching is exciting for geeks but most of the data-analysts spent their time at looking at data and thinking how to represent them. Is there a \"good\" visualization toolkit in Python ? [Matplotlib](http://localhost:8888/notebooks/2_matplotlib.ipynb)\n"
     ]
    },
    {
     "cell_type": "code",
     "collapsed": false,
     "input": [],
     "language": "python",
     "metadata": {},
     "outputs": []
    }
   ],
   "metadata": {}
  }
 ]
}